Accuracy during Training and on NCS not same

[idata]

Folks,

I am having problem deploying a color identifying network on the NCS. From training using Caffe I am getting around 94% accuracy but after generating the graph file and deployed it in NCS for classification it is classifying images wrongly, no where close to 94%. All the layers in the network are supported by NCS. I couldn't figure out where the problem is.

Could someone help us pin point where the problem is, the NCS, or the compile tool, or the network itself? I have attached the network and the deploy prototxt, and also the label text file.

Here is the trained model: https://drive.google.com/file/d/1yTflweZwjL5Gku9X8c5o2zrjbzt0KTlx/view?ts=5aaffa01

Thank you so much in advance.

[idata]

@mulex What are you using to check the NCS result for this network? Just curious. Can you post the log for your NCS result here? I will be debugging your issue and I'll get back to you soon. Thanks.